Not asking for anyone’s secrets of the trade, I’m more curious how people are thinking about context now that newer models chew through huge amounts of it for reasoning. the TLDR: I’m starting to think of context less as working memory and more as a temp scratchpad to start each step. I’m running a
